In vivo spectroscopy and machine learning for the early detection and classification of different stresses in apple trees.

The use of in vivo spectroscopy to detect plant stress in its early stages has the potential to enhance food safety and reduce the need for plant protection products. However, differentiating between various stress types before symptoms appear remains poorly studied. In this study, we investigated the potential of Vis-NIR spectroscopy to differentiate between stress types in apple trees (Malus x domestica Borkh.) exposed to apple scab, waterlogging, and herbicides in a greenhouse. Using a spectroradiometer, we collected spectral signatures of leaves still attached to the tree and utilized machine learning techniques to develop predictive models for detecting stress presence and classifying stress type as early as 1-5 days after exposure. Our findings suggest that changes in spectral reflectance at multiple regions accurately differentiate various types of plant stress on apple trees. Our models were highly accurate (accuracies between 0.94 and 1) when detecting the general presence of stress at an early stage. The wavelengths important for classification relate to photosynthesis via pigment functioning (684 nm) and leaf water (~ 1800-1900 nm), which may be associated with altered gas exchange as a short-term stress response. Overall, our study demonstrates the potential of spectral technology and machine learning for early diagnosis of plant stress, which could lead to reduced environmental burden through optimizing resource utilization in agriculture.